Short-term Prediction Of Festival Passenger Flow Of Urban Rail Transit Based On Adaboost Integrated LSTM

For urban rail transit systems,passenger flow prediction is the basic basis for traffic capacity allocation and operational decision-making.Through the influence of a large number tourists and visiting relatives during festivals,the number of passengers in station during festivals is much higher than that of ordinary passengers.Furthermore,showing strong features of instantaneity and mutability.Moreover,the predicting process has the problem of too little available data.In order to make certain safety operations of the station and increase the organizational efficiency,it is very necessary to research and explore the short-time passenger flow prediction model of during festivals.The paper takes the historical inbound passenger flow series during festivals in four different types of Xi’an Metro stations as samples,and realizes the construction and demonstration of a short-term prediction model of festive passenger flow from optimization of parameters and integration of LSTM.The main contents of this study are as follows:(1)Feature analysis of festival passenger flow and passenger flow data processing.Firstly,from the perspective of time and spatial,the characteristics of passenger flow in Xi’an urban rail transit network,lines and four types of different stations during the festival is analyzed;secondly,this paper explains the definition of short-term passenger flow prediction and related processes,based on the Matlab heat map of passenger flow series The matrix explores the relevance of the station festival passenger flow data,and lays the foundation for data analysis and data processing for festival passenger flow prediction.(2)Establish a prediction model based on LSTM.In view of the non-linear and abrupt characteristics of the station festival inbound passenger flow,the advantage of long and short-term memory(LSTM)neural network is used to process complex time-related series,and the long and short-term memory neural network is used for short-term prediction of station festival inbound passenger flow.The short-term prediction process of inbound passenger flow based on long and short-term memory neural network is constructed,and the parameters are calibrated through conventional empirical method and trial-and-error method to realize experimental simulation.(3)Based on the BGWO-LSTM short-term passenger flow prediction model.In the experiment,it is found that the LSTM network has the disadvantages of large workload and difficult parameter combination determination.The BGWO hybrid intelligent algorithm is designed to optimize the parameter combination of LSTM.The BGWO makes full use of the long-horned beetle algorithm to realize the search with only one long-horned beetle,which can greatly reduce the advantage of the amount of calculation.It integrates the beetle antennae search idea into the gray wolf individual location update strategy,so that the gray wolf individual can move in a better direction.Accurate update,thereby speeding up the convergence speed,improving the optimization performance,and further improving the prediction accuracy.(4)Construct a combined prediction model based on Adaboost integrated BGWO-LSTM.A combination model that combines the efficient optimization ability of the BGWO and the advantages of Adaboost integrated learning "select the best" is proposed.Taking the LSTM model optimized by BGWO as the sub-model,several sub-models are trained through the Adaboost integrated algorithm,and the sub-model weights are determined based on the prediction error,so that the sub-models with small errors have higher weights.Finally,experiments with four different passenger flow series of Xi’an Metro are conducted.The results prove that the combined model can further improve the prediction accuracy,and has greater advantages in the transient and sudden passenger flow series,and the single LSTM model is more suitable for processing the relatively stable passenger flow series.